Automatically recognizing the discourse structure of a body of text

A method in a computer system for generating a discourse structure tree for an input text that characterizes the discourse structure of the input text, the method utilizing syntactic information associated with the input text, the method comprising the steps of:

• identifying in the input text a plurality of clauses;
• for each pair of clauses, determining which of a plurality of possible discourse relations to hypothesize between the pair of clauses based upon the syntactic structure of the input text relative to the pair of clauses;
• and
• applying the hypothesized relations to the clauses to produce a discourse structure tree characterizing the discourse structure of the input text.

A computer-readable medium whose contents cause a computer system to generate a discourse structure tree for a natural language expression that characterizes the discourse structure of the natural language expression, utilizing syntactic information associated with the natural language expression, by performing the steps of:

• selecting in the natural language expression a plurality of clauses;
• for each pair of clauses, determining which of a plurality of possible discourse relations to hypothesize between the pair of clauses based upon the syntactic structure of the natural language expression relative to the pair of clauses;
• and
• applying the hypothesized relations to the clauses to produce a discourse structure tree characterizing the discourse structure of the natural language expression.

A method in a computer system for determining a quantitative score for a discourse structure tree that reflects the likelihood that the discourse structure tree correctly characterizes the discourse structure of a body of text comprised of clauses, the discourse structure tree comprising nodes representing clauses, the nodes including both terminal nodes and non-terminal nodes, the non-terminal nodes each identifying a discourse relation between two or more child nodes, exactly one of the nodes being a root node that is not the child of any node, the method comprising the steps of:

• for each non-terminal node:
• determining a quantitative score for the discourse relation represented by the current node that reflects the likelihood that the discourse relation has been correctly recognized between the clauses represented by the child nodes of the current node,
• combining the scores of any non-terminal child nodes with the score of the discourse relation represented by the current node, and attributing to the current node the combined score;
• and
• attributing to the discourse structure tree the score of the root node.

A computer memory containing a discourse relation hypothesizing data structure for use in hypothesizing discourse relations between a pair of clauses, the data structure comprising, for each of a plurality of relations, a list of one or more entries each containing:

• a list of conditions having an indicated order of application to syntactic information relating to the pair of clauses;
• and
• a quantitative score reflecting the relative likelihood that the relation correctly relates the pair of clauses where the list of conditions is satisfied,
• such that, for the pair of clauses, for one or more of the relations, for each of the entries, the list of conditions may be applied to syntactic information relating to the pair of clauses, and, if the list of conditions is satisfied, the quantitative score may be combined into an overall quantitative score of the relative likelihood that the relation correctly relates the pair of clauses.

A computer memory containing a discourse relation hypothesizing data structure for use in hypothesizing discourse relations between a pair of clauses each comprising a series of words, the data structure comprising, for each of a plurality of relations, a list of one or more entries each containing:

• a list of conditions having an indicated order of application to information available from a lexical knowledge base relating to words within the pair of clauses;
• and
• a quantitative score reflecting the relative likelihood that the relation correctly relates the pair of clauses where the list of conditions is satisfied,
• such that, for the pair of clauses, for one or more of the relations, for each of the entries, the list of conditions may be applied to information relating to words within the pair of clauses procured from a lexical knowledge base, and, if the list of conditions is satisfied, the quantitative score may be combined into an overall quantitative score of the relative likelihood that the relation correctly relates the pair of clauses.

A method in a computer system for recognizing one or more likely discourse relations between a pair of clauses occurring in a body of natural language text, the method comprising the steps of, for each of a plurality of possible relations:

• obtaining for the current possible relation one or more ordered sets of conditions, each condition relating to the syntactic structure of the body of text relative to the pair of clauses;
• and
• for each obtained ordered set of conditions,
• determining, in accordance with the order of the set, whether each of the conditions of the set is satisfied by the syntactic structure of the body of text relative to the pair of clauses, and
• if each of the conditions of the set is satisfied, recognizing the current possible relation as a likely relation between the pair of clauses.

An apparatus for recognizing one or more likely discourse relations between two clauses occurring in a body of natural language text, comprising:

• for each of a plurality of possible relations
• a memory storing, for each of a plurality of possible relations, one or more sets of conditions, each condition relating to the syntactic structure of the body of text relative to the two clauses;
• and
• a likely relation recognition subsystem that, for each of the plurality of possible relations, recognizes the possible relation as a likely relation between the two clauses where each of the conditions of at least one of the sets of conditions is satisfied by the syntactic structure of the body of text relative to the pair of clauses.